This study investigates biometric gait identification using accelerometer and gyroscope signals. The experiments are based on the author’s gait database, which contains information from 100 subjects. Special attention has been paid to techniques for combining extended gait cycles and patterns generated by generative models. These advanced methods allowed for more accurate identification results, thus increasing the F1-score metrics of the biometric system. A convolutional neural network (CNN) was used as the decision model in this study and proved to be effective in recognizing gait patterns. The proposed approach, which combines augmented subsets with generative subsets, significantly improves the accuracy, increasing the F1 score from 0.66 to 0.85.